fix: NQL engine improvements — decimal formatting, schema context, and query accuracy

[William-Hill]

Problems

1. Decimal formatting

Numeric results are returned with excessive decimal places when they should be rendered as percentages. For example 0.75 should display as 75%. This affects columns like pass rate, completion rate, retention probability, etc.

2. Missing / inadequate schema context

The LLM prompt-analyzer does not have reliable schema context, so it guesses column names and table structures. This causes it to query the wrong tables or use non-existent columns.

The schema context needs to include:

• All available tables and their columns with types
• Sample values / value ranges for categorical columns (e.g. grade IN ('A','B','C','D','F','W','I'))
• Which columns are FERPA-sensitive and must never be returned (e.g. student_guid, Student_GUID, PII fields)
• Notes on which table to use for which question type (e.g. course-level DFWI → course_enrollments, student predictions → student_level_with_predictions)

3. Nonsensical query results

For the prompt "What are the top 10 courses with the highest DFW rates?" the engine returned student-level program-of-study data instead of course-level DFWI rates:

Program of Study Year 1  avg completion rate  avg passing rate  total students
120401.00                0.75                 0.76              190
150000.00                0.75                 0.76              186
510000.00                0.75                 0.75              1042
...

Expected result: rows from course_enrollments grouped by course_prefix + course_number with a computed DFWI rate. The LLM queried course_predictions (student-level) and used CIP codes as the "course" identifier.

Root cause: the prompt-analyzer doesn't know course_enrollments exists or how to compute DFWI from the grade column.

Acceptance Criteria

• Numeric results that represent rates/probabilities (values between 0–1) are auto-formatted as percentages in the results table
• prompt-analyzer.ts includes a detailed schema block listing all tables, columns, FERPA-restricted fields, and query guidance
• The prompt "What are the top 10 courses with the highest DFW rates?" returns rows from course_enrollments with correct DFWI rates
• FERPA-sensitive columns (student_guid, name fields, DOB, SSN proxies) are excluded from the schema prompt and blocked from SELECT output

[William-Hill]

Resolved by PR #91.